Intent-Aware Long Short-Term Memory for Intelligent Training of Clinical Handover

Abstract

Clinical handover is a crucial yet high-risk communication event in the provision of safe patient care. However, training standardized clinical handover in real-world scenarios often requires huge labor cost. To tackle with this issue, we propose a computer-aided method for delivering intelligent training of clinical handover at a low labor cost. Specifically, we formulate it as a continuous intent detection task that provides timely feedback during a simulated clinical handover conversation. Towards this goal, we collaborate with experts from a local hospital to collect a clinical handover dataset on real-world handover scenarios. According to the sequential nature of the handover conversation, we further propose the Intent-Aware Long Short-Term Memory (IA-LSTM) model that yields superior performance to baseline methods. Our work shows promise for the computer-aided training of clinical handover in hospitals and can encourage researchers in natural language processing to develop methods on standardized communication.